arXiv:math/9809120v3 [math.

CO] 25 Oct 1998

A combinatorial determinant
Herbert S. Wilf
Department of Mathematics, University of Pennsylvania
Philadelphia, PA 19104-6395

Abstract
A theorem of Mina evaluates the determinant of a matrix with entries Dj (f (x)i ). We note
the important special case where the matrix entries are evaluated at x = 0 and give a simple
proof of it, and some applications. We then give a short proof of the general case.

October 25, 1998

An old theorem of Mina [1], which “deserves to be better known” [3], states that
 !n−1 
 dj 
det f (x)i = 1!2! . . . (n − 1)!f ′ (x)n(n−1)/2 . (1)
 dxj 
i,j=0

A proof of Mina’s theorem can be found, for instance, in [2]. We will first give a short proof of
the special case in which both sides are evaluated at x = 0, with some applications, and then give
a short proof of the general case. The special case shows an interesting structure owing to the
fact that all matrices of that form can be simultaneously triangularized by multiplying them by a
certain universal triangular matrix.

1 Coefficients of powers of power series

Theorem 1 Let f = 1 + a1 x + a2 x2 + . . . be a formal power series, and define a matrix c by1

ci,j = [xj ]f i (i, j ≥ 0). (2)

Then  
n(n+1)/2
det (ci,j )ni,j=0 = a1 (n = 0, 1, 2, . . .). (3)
1
“[xk ]g” means the coefficient of xk in the series g.

1
 To prove this, define a matrix b by
!
i+j i
bi,j = (−1) , (i, j ≥ 0). (4)
j
Then we claim that bc is upper triangular with powers of a1 on its diagonal. Indeed we have
! !
i+j i i j
[xk ]f j = (−1)i [xk ] (−1)j
X X X
bi,j cj,k = (−1) f
j j
j j
j
0, if k < i;

= (−1)i [xk ](1 − f )i = [xk ](a1 x + a2 x2 + . . .)i =
ai1 , if k = i,
as claimed. Since bc is this upper triangular matrix, and det b = 1, the determinant of c is
Qn i n(n+1)/2
i=0 a1 = a1 . ✷
Ed Bender has noted that the hypothesis a0 = 1 can be removed. Indeed if a0 6= 0, apply
the result to f /a0 and discover that the theorem is unchanged. If a0 = 0 the result follows by
continuity.
In order to gain an extra free parameter in the identities that are to follow, as well as to
introduce the idea of the proof of Mina’s theorem in general form, we’ll restate Theorem 1 in terms
of the zth power of f .
Theorem 2 Let f = 1 + a1 x + a2 x2 + . . . be a formal power series, let z be a complex number, and
define a matrix c by
ci,j = [xj ]f zi (i, j ≥ 0). (5)
Then  
det (ci,j )ni,j=0 = (za1 )n(n+1)/2 (n = 0, 1, 2, . . .). (6)

2 Some examples
1. This investigation began when I was looking at the infinite matrix whose (i, j) entry is the
number of representations of the integer j as a sum of i squares of nonnegative integers
(i, j = 0, 1, 2, . . .), and noticed that its determinant is 1. This matrix begins as
 
1 0 0 0 0 0 0 0 ...

 1 1 0 0 1 0 0 0 ... 

1 2 1 0 2 2 0 0 ...
 
 
1 3 3 1 3 6 3 0 ...
 
 
 

 1 4 6 4 5 12 12 4 ... 


 1 5 10 10 10 21 30 20 . . . 

1 6 15 20 21 36 61 60 . . .
 
 

1 7 21 35 42 63 112 141 . . .

 
.. .. .. .. .. .. .. .. . .
 
. . . . . . . . .

2
 This is an array of the type considered in (2) above, where f = 1 + x + x4 + x9 + x16 + . . ..
Hence by Theorem 1, the determinant of every upper-left n × n section of this infinite matrix
is 1. The same will be true if “squares” is replaced by “cubes” or any higher power, or for
that matter by any increasing sequence {0, 1, . . .} at all!

2. Take f = 1 + x in Theorem 2 to discover that

!!n
zi n+1
det = z( 2 ) (n = 0, 1, 2, . . .).
j i,j=0

3. With f = (ex − 1)/x in Theorem 2 we find a determinant that involves the Stirling numbers
of the second kind,
( )!n  n(n+1)/2
(zi)! zi + j z
det = (n = 0, 1, 2, . . .).
(zi + j)! zi i,j=0
2

4. With f = log (1 + x)/x in Theorem 2 we evaluate one that contains the Stirling numbers of
the first kind,
" #!n  n(n+1)/2
(zi)! zi + j z
det = (n = 0, 1, 2, . . .).
(zi + j)! zi i,j=0
2

√
5. Now let f = (1 − 1 − 4x)/(2x) = 1 + x + . . . in Theorem 2. If we use the fact that
√ !k
1− 1 − 4x X k(2j + k − 1)!
= xn ,
2x n≥0
j!(k + j)!

then the resulting determinantal identity can be put in the form

n
(2j + zk − 1)!

n
det = z ( 2 ) 1! 2! 3! . . . (n − 1)! (n = 1, 2, . . .).
(zk + j)! j,k=1

3 The additive structure of this case

This case of Mina’s theorem, in which we evaluate at x = 0, has an interesting structure because,
if the series involved are normalized to constant term 1, then the matrices are all trangularized by
the same matrix b of (4).
Usually, determinants don’t relate well to addition of matrices. That is, if det U , det V can be
evaluated in simple closed form, there is no reason to suppose that the same is true of det (U + V ).
But the determinants that we are now studying act nicely under matrix addition. Thus, if ui,j =

3
[xi ]f j , and vi,j = [xi ]gj , then b(u + v) is triangular, with diagonal entries f ′ (0)i + g′ (0)i (i =
0, 1, 2, . . .), and so
n
det (ui,j + vi,j )ni,j=0 = (f ′ (0)i + g ′ (0)i ).
Y

i=0
For example, !!n n
!
ri si
(r i + si ),
Y
det + =
j j i,j=0 i=0

and more generally if S is any set of numbers, and c is a function on S, then

!!n n
!
X ri Y X
j
det c(r) = c(r)r .
r∈S
j j=0 r∈S
i,j=0

An interesting special case is obtained by taking S to be a set of m equally spaced points in an

interval, say (0, 1), taking the c(r)’s all equal to 1/m, and taking the limit as m → ∞. The result
is that ! !n
Z 1 n
xi Y
det f (x) dx = µj (f ),
0 j i,j=0 j=0
R1
where the µi (f ) = 0 xi f (x)dx are the moments of f . For example,
! !n
Z 1 xi 1
det dx = .
0 j i,j=0
(n + 1)!

4 Proof of Mina’s theorem

Now we prove Mina’s theorem by proving the following small generalization of it.
 !n 
 dj  Pn
n(n+1)/2
f (x)zxi = f (x)z (xi −i)
Y
det 0 zf ′ (x) (xj − xi ). (7)
 dxj 
i,j=0 0≤i<j≤n

This will follow easily from the following observation.

Lemma 1 Let {pj (x)} (j = 0, 1, 2, ...) be any sequence of polynomials such that the jth one is of
degree j, for each j ≥ 0. Then the determinant of the matrix Q = {pj (xi )}ni,j=0 is equal to the
product of the highest coefficients of p0 , p1 , . . . , pn times the discriminant of the xi ’s.
Indeed, if A is the lower triangular matrix of coefficients of the pj ’s and V is the Vandermonde
matrix of the xi ’s then Q = AV . ✷
Now let αi,j = D j (f (x)zxi ), where D = d/dx. We claim that, for each j = 0, 1, . . . , αi,j is of
the form f zxi −j times a polynomial in z of exact degree j, whose coefficients are functions of x

4
and whose highest coefficient is (xi f ′ )j . In view of Lemma 1, this claim will prove the theorem.
However this claim is trivial to prove by induction on j. ✷
Theorem 2 is the special case of (7) in which x = 0, z = 1, and the polynomials pj (x) are given
by
j
xℓ j  
[t ] log y(t)ℓ ,
X
pj (x) =
ℓ=0
ℓ!

with highest coefficients aj1 /j!.

5 Remarks
1. The family of matrices of the form (2) was previously investigated by Doron Zeilberger [4],
from the point of view of constant term identities.

2. My thanks go to Brendan McKay for some e-mail exchanges that were helpful in clarifying
the ideas here.

3. By comparing (1) and (7) we see that the absence of f (x) on the right side of the former is
because of the “lucky coincidence” that in the former, x0 + x1 + . . . + xn = n(n + 1)/2.

References
[1] L. Mina, Formole generali delle derivate successive d’una funzione espresse mediante quelle
della sua inverse, Giornale di Mat. xliii (1904), 196-212.

[2] Volker Strehl and Herbert S. Wilf, Five surprisingly simple complexities, Symbolic compu-
tation in combinatorics ∆1 (Ithaca, NY, 1993), J. Symbolic Comput. 20 (1995), no. 5-6,
725–729.

[3] A. J. van der Poorten, Some determinants that should be better known, J. Austral. Math.
Soc. 21 (Series A) (1976), 278-288.

[4] Doron Zeilberger, A constant term identity featuring the ubiquitous (and mysterious)
Andrews-Mills-Robbins-Rumsey numbers {1, 2, 7, 42, 429, ...}, J. Combinatorial Theory (Ser.
A) 66 (1994), 17–27.